Approximate Exploitability: Learning A Best Response In Large Games

Abstract

Researchers have demonstrated that neural networks are vulnerable to adversarial examples and subtle environment changes, both of which one can view as a form of distribution shift. To humans, the resulting errors can look like blunders, eroding trust in these agents. In prior games research, agent evaluation often focused on the in-practice game outcomes. While valuable, such evaluation typically fails to evaluate robustness to worst-case outcomes. Prior research in computer poker has examined how to assess such worst-case performance, both exactly and approximately. Unfortunately, exact computation is infeasible with larger domains, and existing approximations rely on poker-specific knowledge. We introduce ISMCTS-BR, a scalable search-based deep reinforcement learning algorithm for learning a best response to an agent, thereby approximating worst-case performance. We demonstrate the technique in several two-player zero-sum games against a variety of agents, including several AlphaZer

Stats

• citations0
• S2 citations—
• github stars0
• HF likes0
• heat score0.00
• arxiv keytimbers2020approximate

Related papers

• Efficient Exploration Of Zero-sum Stochastic Games (2020)0.00
• Learning From Learners: Adapting Reinforcement Learning Agents To Be Competitive In A Card Game (2020)0.00
• In-context Exploiter For Extensive-form Games (2024)0.00
• Combining Tree-search, Generative Models, And Nash Bargaining Concepts In Game-theoretic Reinforcement Learning (2023)0.00
• Learning To Play No-press Diplomacy With Best Response Policy Iteration (2020)0.00
• Are Alphazero-like Agents Robust To Adversarial Perturbations? (2022)0.00
• A Deep Reinforcement Learning Approach For Finding Non-exploitable Strategies In Two-player Atari Games (2022)0.00
• Evaluation And Learning In Two-player Symmetric Games Via Best And Better Responses (2022)0.00